WiFi Channel State Information-based Motion Detection Across Time-Domain

Human presence is a basic requirement for realizing other wireless sensing tasks. With the development of wireless communication technology, WiFi is not only used as a widely used network access method, but also applied in various intelligent environment monitoring and human behavior recognition fields. The motion human detection technique based on WiFi Channel State Information (CSI) is becoming a research hotspot due to its non-invasive and high-precision advantages. However, most of the existing studies are limited to mixing and analyzing data collected within a short period of time, with less research in detection across time-domain. Because CSI data are subject to random changes due to temporal factors, this makes it difficult to accurately predict future data using the used data patterns. For this reason, we propose algorithms based on motion human detection across time-domain. We first perform outlier removal for CSI amplitude and then filter the signal using wavelet denoising. In order to improve the accuracy of motion human detection across time-domain, we use the average variance of sliding time window as the robust feature of motion human and combine it with other statistical features to build the feature set. Finally, an integrated machine learning model is utilized in order to determine the future indoor human states. Experimental results show that our method achieves an average accuracy of more than 98% in detecting motion human across time-domain.